LLM: update qlora example (#9454)

* update qlora example

* fix loss=0

python/llm/example/GPU/QLoRA-FineTuning/README.md

@@ -17,7 +17,7 @@ conda activate llm
 # below command will install intel_extension_for_pytorch==2.0.110+xpu as default
 # you can install specific ipex/torch version for your need
 pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
-pip install transformers==4.34.0
+pip install datasets transformers==4.34.0
 pip install peft==0.5.0
 pip install accelerate==0.23.0
 ```

python/llm/example/GPU/QLoRA-FineTuning/qlora_finetuning.py

@@ -48,7 +48,9 @@ if __name__ == "__main__":
                                                 torch_dtype=torch.float16,
                                                 modules_to_not_convert=["lm_head"],)
     model = model.to('xpu')
-    model.gradient_checkpointing_enable()
+    # Enable gradient_checkpointing if your memory is not enough,
+    # it will slowdown the training speed
+    # model.gradient_checkpointing_enable()
     model = prepare_model_for_kbit_training(model)
     config = LoraConfig(
         r=8, 
@@ -69,9 +71,10 @@ if __name__ == "__main__":
             gradient_accumulation_steps= 1,
             warmup_steps=20,
             max_steps=200,
-            learning_rate=2e-4,
+            learning_rate=2e-5,
             save_steps=100,
-            fp16=True,
+            # fp16=True,
+            bf16=True,  # bf16 is more stable in training
             logging_steps=20,
             output_dir="outputs",
             optim="adamw_hf", # paged_adamw_8bit is not supported yet